A deep learning approach to depression detection based on Convolutional Neural Networks and Transfer Learning

Abstract

Depression and mental health issues (stress, nervousness, panic attacks, anxiety attacks etc.) are nowadays a major issue in the whole world. It is a common cause of mental illness that has been linked to an increased risk of dying young. Especially in our country, mental health is an issue which most of the families do not want to give as much attention as it is supposed to get and because of that so many people who are suffering from Major Depressive Disorder (MDD) are often helpless. Currently there are numerous ways to detect depression by various methods. For example: emotion recognition, social media records, analyzing daily routine with the help of machine learning and many more. This paper aims to detect depression by implementing various deep learning/ transfer learning models (for example: VGG16, Xception, ResNet152, MobileNetV2 etc.) using EEG brain signals to discover the model that provides the highest level of accuracy for our data type. In addition, we want to analyze why the particular model performs better and what might be the cases to make a model perform better to propose a model so that this method of modeling can be used in most cases for detecting depression and model improvement. Furthermore, we have made a custom model which gives the most accuracy (99.75%). We are successful at bringing the highest accuracy among the existing models which were implemented by us. For this reason, we are analyzing the EEG brain signal data of several healthy and MDD patients. We believe that this research will aid in the development of innovative strategies for building models and early identification of depression in our daily lives.